NATTEN
Copyright (c) 2022 - 2026 Ali Hassani.
Refer to LICENSE for details.

This project's CUDA C++ kernels are largely based on kernels from the NVIDIA CUTLASS Project.
CUTLASS is included as a submodule under third_party/cutlass/, and can also be accessed at
https://github.com/NVIDIA/cutlass/, available under the BSD 3-Clause License (see
third_party/cutlass/LICENSE.txt).

Ampere Fused Neighborhood Attention kernels are based on the memory-efficient
attention kernels from the xFormers project by Meta Platforms, Inc, which is available at
https://github.com/facebookresearch/xformers/ under the BSD 3-Clause License (see
https://raw.githubusercontent.com/facebookresearch/xformers/refs/heads/main/LICENSE).

csrc/include/natten/cuda/fmha and csrc/include/natten/cuda/fna are mostly based on the xFormers
CUTLASS attention kernels.
